Obtaining fast error rates in nonconvex situations

Shahar Mendelson

doi:10.1016/j.jco.2007.09.001

Obtaining fast error rates in nonconvex situations

Shahar Mendelson^*

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

18 Citations (Scopus)

Abstract

We show that under mild assumptions on the learning problem, one can obtain a fast error rate for every reasonable fixed target function even if the base class is not convex. To that end, we show that in such cases the excess loss class satisfies a Bernstein type condition.

Original language	English
Pages (from-to)	380-397
Number of pages	18
Journal	Journal of Complexity
Volume	24
Issue number	3
DOIs	https://doi.org/10.1016/j.jco.2007.09.001
Publication status	Published - Jun 2008

Access to Document

10.1016/j.jco.2007.09.001

Cite this

@article{68a6fb822e204ea28443364fcd793ff0,

title = "Obtaining fast error rates in nonconvex situations",

abstract = "We show that under mild assumptions on the learning problem, one can obtain a fast error rate for every reasonable fixed target function even if the base class is not convex. To that end, we show that in such cases the excess loss class satisfies a Bernstein type condition.",

keywords = "Bernstein condition, Nearest point map, Nonconvex classes",

author = "Shahar Mendelson",

year = "2008",

month = jun,

doi = "10.1016/j.jco.2007.09.001",

language = "English",

volume = "24",

pages = "380--397",

journal = "Journal of Complexity",

issn = "0885-064X",

publisher = "Academic Press Inc.",

number = "3",

}

TY - JOUR

T1 - Obtaining fast error rates in nonconvex situations

AU - Mendelson, Shahar

PY - 2008/6

Y1 - 2008/6

N2 - We show that under mild assumptions on the learning problem, one can obtain a fast error rate for every reasonable fixed target function even if the base class is not convex. To that end, we show that in such cases the excess loss class satisfies a Bernstein type condition.

AB - We show that under mild assumptions on the learning problem, one can obtain a fast error rate for every reasonable fixed target function even if the base class is not convex. To that end, we show that in such cases the excess loss class satisfies a Bernstein type condition.

KW - Bernstein condition

KW - Nearest point map

KW - Nonconvex classes

UR - http://www.scopus.com/inward/record.url?scp=44649102735&partnerID=8YFLogxK

U2 - 10.1016/j.jco.2007.09.001

DO - 10.1016/j.jco.2007.09.001

M3 - Article

SN - 0885-064X

VL - 24

SP - 380

EP - 397

JO - Journal of Complexity

JF - Journal of Complexity

IS - 3

ER -

Obtaining fast error rates in nonconvex situations

Abstract

Access to Document

Other files and links

Fingerprint

Cite this